The impact of COVID-19 pandemic on influenza transmission: molecular and epidemiological evidence

To quantify the impact of COVID-19-related control measures on the spread of human influenza virus, we analyzed case numbers, viral molecular sequences, personal behavior data, and policy stringency data from various countries, and found consistent evidence of decrease in influenza incidence after the emergence of COVID-19.


Introduction
The emergence and spread of COVID-19 in 2020 led to a number of large-scale public health measures to limit international travel, reduce gatherings, and increase mask-wearing. While these preventative measures were implemented to curtail the COVID-19 pandemic, they seem to have also impacted the spread of other respiratory illnesses. There have been several reports on the decrease in case numbers during 2019-2020 influenza season in the northern hemisphere [6], and the lack of a 2020 influenza season in the southern hemisphere [8]. Here, we quantify the impact of the COVID-19 pandemic on the spread of influenza in terms of incidence and viral molecular diversity [1].

Case count data
We analyzed weekly case count data of influenza available in FluNet [10] from various regions during the 2010-2020 influenza seasons. We defined TS and TE as the weeks during which the estimated number of cases reached 10% and 90% of the total case numbers in each influenza . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint season, respectively. Since the influenza outbreaks for most regions started before the COVID-19 outbreak, we compared TE and durations of influenza seasons pre-and post-COVID-19 pandemic. We defined the duration of an influenza season by the difference between TS and TE, and standardized the duration in the 2019-2020 season by the average and standard deviation of the duration from previous 9 seasons.

Viral molecular sequences
We analyzed the HA segment of human influenza A H1N1 and H3N2 sequences available in the GISAID EpiFlu database on November 1 st 2020. The collection dates of the sequences ranged from January 2016 to December 2020. We used BEAST [3] to estimate the effective population size (Ne) from 2016 to 2020 for each location. The numbers of sequences analyzed are indicated in Appendix Table 1.

Results and Discussion
To examine the indirect impact of COVID-19 on influenza dynamics, we compared the 2019-2020 influenza season with previous 9 seasons in 21 locations across 5 continents (Table 1). We found that for all locations in Asia, the outbreak in 2019-2020 both ended earlier and lasted for a shorter duration than previous years. In the Americas, Europe, and Africa, 12 locations out of 15 showed an earlier end of the flu season in the 2019-2020 season than previous years; the rest remained similar. For locations where influenza seasons usually end later in the year, such as Brazil, Guatemala, and South Africa, the difference in duration between the 2019-2020 season and previous seasons was larger than other locations. The flu season in the Southern Hemisphere, which usually starts much later in the year, disappeared in several countries in 2020 [6,8].
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint In addition to case count data, we analyzed molecular data to evidence our findings. For each location and each influenza type, we calculated the within-location genetic diversity, Watterson's θ [9], for the first half of 2019 and for the first of 2020 (Appendix Table 2). We found that for 10 out of 11 locations we analyzed, θ decreased from 2019 to 2020 for H1N1, and 9 out of 11 locations for H3N2. On the other hand, we calculated between-region genetic diversity for each pair of regions once in 2019 and again in 2020; 9 out of 11 locations for H1N1 and 6 out of 11 locations for H3N2 had their between-region genetic diversity increase from 2019 to 2020 in at least 50% of pairs for which the location was involved, reflecting reduced travel between regions in 2020 (Appendix Table 3) .
While Watterson's θ measures overall viral diversity, the effective population size (Ne) quantifies genetic diversity over time [4]. We estimated Ne for H1N1 in 11 countries and found a decrease in Ne in 9 countries, including Italy and Taiwan ( Table 1).
For personal measures and government policies against COVID-19, we also noticed that Asian countries tend to act earlier than countries in other continents, especially wearing masks (Appendix Table 4). Taken together, we observed earlier ends of flu seasons in Asia than in Europe and America, which could be explained by the earlier implementation of nonpharmaceutical interventions.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint The decrease in influenza incidence after the emergence of COVID-19 was confirmed by both case count and molecular data, providing stronger support for a genuine decrease in influenza incidence, even when count data might be incomplete or imprecise due to potentially less flu surveillance in 2020.

Acknowledgements
We thank GISAID and sequencing laboratories for making the influenza A viral molecular sequences available Specific laboratories and accession numbers are found at the end of the Appendix. HHC and DWH were supported by the Ministry of Science and Technology in Taiwan (MOST 110-2636-B-007-009); HHC was supported by the Yushan Scholar Program.
The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

First Author Biography
Leon King Tran is an undergraduate student in Mathematics and master's student in Statistics at Stanford University and is interested in Biostatistics and Stochastic Processes.
Dai-Wei Huang is an undergraduate student in the Interdisciplinary Program of Life Science, National Tsing Hua University, Hsinchu, Taiwan and is currently doing research related to epidemiology and biostatistics.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Figure 1. Effective Population Size Trajectory of H1N1 in Italy and H3N2 in Taiwan
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Personal measures and governmental policies against COVID-19
We analyzed two personal measures taken against the spread of COVID-19 from YouGov [7]wearing masks when in public places and improving personal hygiene, from February 21, 2020 to December 17, 2020. For governmental responses to COVID-19, we used the data from OxCGRT, the Oxford COVID-19 Government Response Tracker [5], from January 1, 2020 to May 31, 2020.

Molecular Epidemiology
We estimated a mutation rate of H3N2 for New York State of 4 x 10 -3 per base per year and 3.6 x 10 -3 for H1N1. We then fixed the corresponding global mutation rates for the rest of the locations. We assumed the HKY prior mutation model with empirical frequencies and Gamma+Invariant site heterogeneity with 4 Gamma categories and 3 partitions into codon positions and assumed the Bayesian Skyride Coalescent prior on effective population size trajectories. We examined convergence by the likelihood trace plot and the effective sample size for each location and type. To estimate Ne, we divided the output (Ne * Gt) by the generation time (Gt) for H1N1 (2.3 days) and for H3N2 (3.1 days) [2]. We ran the MCMC for 10,000,000 iterations, thinning every 1000 iterations and with 10% of burnin. We removed any regions without convergence.
The within-region diversities were measured by Watterson's θ. The between-region diversities were measured as follows: Between-region diversity between region 1 and region 2 = . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint Appendix Table 1. Effective sample sizes of root age and number of sequences per country for influenza types H1N1 and H3N2. These effective sample size numbers were obtained with Tracer [3]. Effective sample sizes with N/A indicate lack of convergence and were omitted from the analysis. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint Appendix Table 2. Within-region genetic diversity during the first half of 2019 compared to the first half of 2020, for influenza types H1N1 and H3N2. Diversity was measured by Watterson's θ. The decrease column indicates whether genetic diversity decreased from the first half of 2019 to the first half of 2020. Sequences with more than 5% gaps were removed from the analysis. N/A indicates a lack of data. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint Appendix Table 3. Between-region diversity during the first half of 2019 compared to the first half of 2020, for influenza types H1N1 and H3N2. For each pair of regions, we calculated the between-region diversity in the first half of 2019 and in the first half of 2020. The third column for each region denotes the number of pairs involving that region with an increase in betweenregion diversity from 2019 to 2020, divided by the total number of pairwise comparisons. Regions where this fraction is above 0.5 are marked in grey. N/A indicates that there was not enough data to calculate diversity. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 10, 2021. ; https://doi.org/10.1101/2021.06.08.21258434 doi: medRxiv preprint